The application of a learning signal processing system to blood velocity measurement is described. The system has been developed to increase the accuracy of the measurement at low S/N (signal-to-noise) ratio. An artificial neural network learns the pulsatile fluctuation of the velocity in order to predict the succeeding velocity signals adaptively. The velocity of red blood cells (RBCs) in microvessels of rat mesentery was measured by using a microscopic laser Doppler velocimeter. Cardiac pulsation apparently affects the RBC velocity even in arterioles; however, detection of the pulsatile component in the RBC velocity fluctuation using the conventional signal processor is very difficult because of its low S/N ratio. In contrast, the learning signal processing system which is pretrained can detect the component at low S/N ratio.